Experimental low latency gaming build

Everything about latency. This section is mainly user/consumer discussion. (Peer-reviewed scientific discussion should go in Laboratory section). Tips, mouse lag, display lag, game engine lag, network lag, whole input lag chain, VSYNC OFF vs VSYNC ON, and more! Input Lag Articles on Blur Busters.
User avatar
axaro1
Posts: 628
Joined: 23 Apr 2020, 12:00
Location: Milan, Italy

Re: Experimental low latency gaming build

Post by axaro1 » 23 Jul 2023, 04:26

PunksOnHope wrote:
22 Jul 2023, 11:54
101 base clock causes cpu to kill turbo functionality and locks it to 3.43 mhz. I guess if u do base clock u have to adjust multipliers? I thought they were locked for x3d
My CPU behaviour is different: when I launch a cpu intensive game with HWinfo open I see all my cores jumps to 4539mhz (maybe because I disabled c-states?). I also get better ST and MT results with this setup compared to 100mhz BCLK.

It also never drop down to base clock (which should be 3.4*1.02=3.468). Sometimes 1 core drops to slightly above 3600mhz when I'm not 100% saturating the chip or when idling but when gaming, stress testing and benchmarking I see all my cores reach 4539mhz.
I know that people on overclock.net forum argue that their chip can't boost with C-States disabled and/or custom BCLK but in practice it doesn't appear to be the case with my motherboard (I have a Gigabyte AB350 Gaming 3).

Edit:
Here's an example playing Overwatch with uncapped fps (still not maxing out the cpu):
Image
Last edited by axaro1 on 23 Jul 2023, 07:01, edited 2 times in total.
XL2566K* | XV252QF* | LG C1* | HP OMEN X 25 | XL2546K | VG259QM | XG2402 | LS24F350[RIP]
*= currently owned



MONITOR: XL2566K custom VT: https://i.imgur.com/ylYkuLf.png
CPU: 5800x3d 102mhz BCLK
GPU: 3080FE undervolted
RAM: https://i.imgur.com/iwmraZB.png
MOUSE: Endgame Gear OP1 8k
KEYBOARD: Wooting 60he

User avatar
axaro1
Posts: 628
Joined: 23 Apr 2020, 12:00
Location: Milan, Italy

Re: Experimental low latency gaming build

Post by axaro1 » 23 Jul 2023, 04:59

LAGenjoyer666 wrote:
22 Jul 2023, 09:01
  • C-States DISABLED (c-states can affect USB behaviour on AM4 motherboards, it's honestly better to disable it most of the times)
  • Cool&Quiet DISABLED
  • CPPC ENABLED
  • CPPC Preferred Core DISABLED (this is a must with x3d chips for better workload allocations)
  • PCIe ASPM mode DISABLED (remove PCIE power saving)
  • L1/L2 Stream HW Prefetcher ENABLED
SMU Common Options:
  • APBDIS 1 with Fixed SOC Pstate 0 (forces infinity fabric and memory controller to run at full power)
  • DF Cstates DISABLED (this way you your infinity fabric never run into idle lower power states)
  • NBIO LCLK DPM -> NBIO DPM Control set it Manual to 2 (forces your socket to always run at the highest power states, it doesn't only affect PCIe but it affects the entire socket)
Do you think these are good settings for 5900x too?
You can 100% copu SMU Common Options but don't touch CPPC settings because these are CPU specific and the 5800x3d is know for performing better with Preferred Core disabled.
XL2566K* | XV252QF* | LG C1* | HP OMEN X 25 | XL2546K | VG259QM | XG2402 | LS24F350[RIP]
*= currently owned



MONITOR: XL2566K custom VT: https://i.imgur.com/ylYkuLf.png
CPU: 5800x3d 102mhz BCLK
GPU: 3080FE undervolted
RAM: https://i.imgur.com/iwmraZB.png
MOUSE: Endgame Gear OP1 8k
KEYBOARD: Wooting 60he

User avatar
axaro1
Posts: 628
Joined: 23 Apr 2020, 12:00
Location: Milan, Italy

Re: Experimental low latency gaming build

Post by axaro1 » 23 Jul 2023, 05:07

PunksOnHope wrote:
22 Jul 2023, 10:58
I'm currently swapping out memory sticks because I feel it's unstable with 4 dimms. Do u think above 3200mhz matters for vcache?
Yes it does, not as much as with non 3d chips, but some games that barely scale with cache, such as CSGO, will see noticeable improvements from RAM oc.

In case of misses and generally low cache hit rate scenarios the chip will always default back to accessing RAM.
XL2566K* | XV252QF* | LG C1* | HP OMEN X 25 | XL2546K | VG259QM | XG2402 | LS24F350[RIP]
*= currently owned



MONITOR: XL2566K custom VT: https://i.imgur.com/ylYkuLf.png
CPU: 5800x3d 102mhz BCLK
GPU: 3080FE undervolted
RAM: https://i.imgur.com/iwmraZB.png
MOUSE: Endgame Gear OP1 8k
KEYBOARD: Wooting 60he

deama
Posts: 371
Joined: 07 Aug 2019, 12:00

Re: Experimental low latency gaming build

Post by deama » 23 Jul 2023, 07:38

I get slightly better input lag with my 7950x3dv than my old 3600.
I measured using a high speed camera.

PunksOnHope
Posts: 36
Joined: 18 Jan 2023, 07:47

Re: Experimental low latency gaming build

Post by PunksOnHope » 23 Jul 2023, 21:40

Another update it seems my memory controller and mobo were having issues running 4 dimms of SR which is unfortunate because now I don't have the equivalent of DR with only 2 dimms. Microstutters have gone down but are still present.

What I did find to have a better effect is swapping from display port to hdmi. It forces it to 8 bit color which may have something to do with it. Freesync seems to not work well with my viewsonic vg2431 monitor and hdmi especially this cheap hdmi cable.

What I recently tried was using amd software suite (6700xt) I locked my clock to 2400/2500. Any kind of overclock or undervolt would cause my card to fail and reset driver settings after a day or two. So far locked at this frequency is stable with stock settings and just locking the clock below boost.

I then experimented with freesync and enhanced sync + antilag. This helped some for big stutter but introduced microstutter as in the monitor osd something is bugging out it shows fluctuations from 140hz and jumping sporadically all the way to 190 but never 240 which I should be getting in the 300 to 400s in paladins and overwatch. I then turned off freesync and now im running with antilag + esync and rsr in driver. This seems to be the most smooth settings for now. The latency hit from the upscale is overcame by raw frames. 1600x900 upscaled offers about a 20% increase in framerate and the sharpening effect makes games that I play look noticeably better. Games like borderlands, overwatch, paladins, apex. My only qualms with this is I wonder how much this effects pixel precision when aiming in a sense I highly doubt it. But so far this still has noticeable input lag and frametime issues but it's far better than stock settings.

On a side note is ilsc worth running? Or does the monitoring aspect of it incur more of a penalty than any possible savings.

PunksOnHope
Posts: 36
Joined: 18 Jan 2023, 07:47

Re: Experimental low latency gaming build

Post by PunksOnHope » 23 Jul 2023, 21:44

I really wish amd would release a hags driver as the windows version with it is nearly 4 months old now and lacks all of the features from the amd release. Hags driver showed promise but it ran worse in many games and had issues with microstutters but when it worked it worked well.

On a side note I stepped away from gaming with anything above 1k polling rate mice. My viper 8k still clicks in 8k but I noticed my mouse inputs feel accelerated with 8khz mice and it's terrible for muscle memory to me. When I'm doing a lot of fast flicking regularly it feels good when I'm doing a lot of wrist aiming and microadjustments they feel noticeably slower than my edge of screen flicks and it's just not worth the mental desync.

babooz99
Posts: 62
Joined: 14 Mar 2023, 11:12

Re: Experimental low latency gaming build

Post by babooz99 » 24 Jul 2023, 02:56

Hmm my 7800x3d feels more responsive than 12700kf at same fps.

User avatar
imprecise
Posts: 287
Joined: 16 Nov 2022, 13:47

Re: Experimental low latency gaming build

Post by imprecise » 02 Aug 2023, 19:25

PunksOnHope wrote:
22 Jul 2023, 10:58

I'm currently swapping out memory sticks because I feel it's unstable with 4 dimms. Do u think above 3200mhz matters for vcache?
EDIT: I had infinity fabric range at 3800-4000 but it should be 3600-4000.

I found running a 5950x with 3200mhz RAM speed turned out to be causing slowness. XMP settings on my BIOS automatically choose speeds like 3200-3400mhz, which was also causing slowness. I looked into infinity fabric speeds and found the RAM needs to be running at 3600-4000 without adjusting any ratios. I think if you try this, you will find a big improvement. The numbers might be listed in the BIOS at half the speed (1800mhz-2000mhz). I spent a lot of time adjusting timings and such at the lower frequencies but performance at 3600-4000 with really slow timings is just better, and it's a known issue for the AMD infinityfabric design. Running below 3600 creates a bottleneck that was not part of the intended use.

As for optimizing for input lag so the system doesn't perform worse than an old low-spec computer, I would recommend optimizing BIOS like suggested here and optimize Windows for known input lag issues. Each new version of Windows has created added lag due to new features, but there are still ways to turn these off. HPET timer is also defective with AMD guidance to disable it since 2018. After disabling core parking and HPET, and lowering priority of certain apps, my system feels fully responsive on W11. I also had a lot of issues with this PC and observed it was underperforming much older systems, and had similar issues with 2 other systems from both intel and AMD.

Post Reply